A recent kernel change caused the previously unused atomic_cmpset_masked() to
be used.
It had a typo in it.
Instead of reading the old value from an uninitialized variable, read it
from the passed-in pointer as intended.
This fixes crashes on 64 bit Book-E.
Obtained from: jhibbits
The goal of this change is to make the atomic_load_acq_{8,16},
atomic_testandset{,_acq}_long, and atomic_testandclear_long primitives
available in MI-namespace.
The second goal is to get this draft out of my local tree, as anything that
requires a full tinderbox is a big burden out of tree. MD specifics can be
refined individually afterwards.
The generic implementations may not be ideal for your architecture; feel
free to implement better versions. If no subword_atomic definitions are
needed, the include can be removed from your arch's machine/atomic.h.
Generic definitions are guarded by defined macros of the same name. To
avoid picking up conflicting generic definitions, some macro defines are
added to various MD machine/atomic.h to register an existing implementation.
Include _atomic_subword.h in arm and arm64 machine/atomic.h.
For some odd reason, KCSAN only generates some versions of primitives.
Generate the _acq variants of atomic_load.*_8, atomic_load.*_16, and
atomic_testandset.*_long. There are other questionably disabled primitives,
but I didn't run into them, so I left them alone. KCSAN is only built for
amd64 in tinderbox for now.
Add atomic_subword implementations of atomic_load_acq_{8,16} implemented
using masking and atomic_load_acq_32.
Add generic atomic_subword implementations of atomic_testandset_long(),
atomic_testandclear_long(), and atomic_testandset_acq_long(), using
atomic_fcmpset_long() and atomic_fcmpset_acq_long().
On x86, add atomic_testandset_acq_long as an alias for
atomic_testandset_long.
Reviewed by: kevans, rlibby (previous versions both)
Differential Revision: https://reviews.freebsd.org/D22963
This is a lock-based emulation of 64-bit atomics for kernel use, split off
from an earlier patch by jhibbits.
This is needed to unblock future improvements that reduce the need for
locking on 64-bit platforms by using atomic updates.
The implementation allows for future integration with userland atomic64,
but as that implies going through sysarch for every use, the current
status quo of userland doing its own locking may be for the best.
Submitted by: jhibbits (original patch), kevans (mips bits)
Reviewed by: jhibbits, jeff, kevans
Differential Revision: https://reviews.freebsd.org/D22976
Summary:
This is a more optimal way of doing atomic_compset_masked() than the
fallback in sys/_atomic_subword.h. There's also an override for
_atomic_fcmpset_masked_word(), which may or may not be necessary, and is
unused for powerpc.
Reviewed by: kevans, kib
Differential Revision: https://reviews.freebsd.org/D22359
|
This adds two implementations for each atomic_fcmpset_ and atomic_cmpset_
short and char functions, selectable at compile time for the target
architecture. By default, it uses a generic shift-and-mask to perform atomic
updates to sub-components of 32-bit words from <sys/_atomic_subword.h>.
However, if ISA_206_ATOMICS is defined it uses the ll/sc instructions for
halfword and bytes, introduced in PowerISA 2.06. These instructions are
supported by all IBM processors from POWER7 on, as well as the Freescale/NXP
e6500 core. Although the e5500 and e500mc both implement PowerISA 2.06 they
do not implement these instructions.
As part of this, clean up the atomic_(f)cmpset_acq and _rel wrappers, by
using macros to reduce code duplication.
ISA_206_ATOMICS requires clang or newer binutils (2.20 or later).
Differential Revision: https://reviews.freebsd.org/D21682
Both IBM and Freescale programming examples presume the cmpset operands will
favor equal, and pessimize the non-equal case instead. Do the same for
atomic_cmpset_* and atomic_fcmpset_*. This slightly pessimizes the failure
case, in favor of the success case.
MFC after: 3 weeks
'sync' is pretty heavy-handed, and is unnecessary for this use case. It's a
full barrier, which is applicable for all storage types. However,
atomic_load_acq_*() is only expected to operate on physical memory, not
device memory, so lwsync is sufficient (lwsync provides access ordering on
memory that is marked as Coherency Required and is not Write Through nor
Cache Inhibited). On 32-bit systems, this is a nop, since powerpc_lwsync()
is defined to use sync, as a workaround for a silicon bug in the Freescale
e500 core.
They provide relaxed-ordered atomic access semantic. Due to the
FreeBSD memory model, the operations are syntaxical wrappers around
the volatile accesses. The volatile qualifier is used to ensure that
the access not optimized out and in turn depends on the volatile
semantic as implemented by supported compilers.
The motivation for adding the operation is to help people coming from
other systems or knowing the C11/C++ standards where atomics have
special type and require use of the special access operations. It is
still the case that FreeBSD requires plain load and stores of aligned
integer types to be atomic.
Suggested by: jhb
Reviewed by: alc, jhb
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D13534
Mainly focus on files that use BSD 2-Clause license, however the tool I
was using misidentified many licenses so this was mostly a manual - error
prone - task.
The Software Package Data Exchange (SPDX) group provides a specification
to make it easier for automated tools to detect and summarize well known
opensource licenses. We are gradually adopting the specification, noting
that the tags are considered only advisory and do not, in any way,
superceed or replace the license texts.
The desired behavior of atomic_fcmpset_() is to always exit on error. Instead
of retrying on lost reservation, leave the retry to the caller, and return
error.
Reported by: kib
There are no alternatives defined, so there's no point in keeping them. Also,
they weren't around every inline asm block anyway. Without __GNUCLIKE_ASM
defined, the guarded functions return garbage.
Reported by: Andrew Thompson
Summary:
atomic_fcmpset_*() is analogous to atomic_cmpset(), but saves off the read value
from the target memory location into the 'old' pointer in the case of failure.
Requested by: mjg
Differential Revision: https://reviews.freebsd.org/D9325
lwsync instruction, which does not provide Store/Load barrier. Fix
this by using "full" sync barrier for mb().
atomic_store_rel() does not need full barrier, change mb() call there
to the lwsync instruction if not hitting the known CPU erratas
(i.e. on 32bit). Provide powerpc_lwsync() helper to isolate the
lwsync/sync compile time selection, and use it in atomic_store_rel()
and several other places which duplicate the code.
Noted by: alc
Reviewed and tested by: nwhitehorn
Sponsored by: The FreeBSD Foundation
provide a semantic defined by the C11 fences with corresponding
memory_order.
atomic_thread_fence_acq() gives r | r, w, where r and w are read and
write accesses, and | denotes the fence itself.
atomic_thread_fence_rel() is r, w | w.
atomic_thread_fence_acq_rel() is the combination of the acquire and
release in single operation. Note that reads after the acq+rel fence
could be made visible before writes preceeding the fence.
atomic_thread_fence_seq_cst() orders all accesses before/after the
fence, and the fence itself is globally ordered against other
sequentially consistent atomic operations.
Reviewed by: alc
Discussed with: bde
Sponsored by: The FreeBSD Foundation
MFC after: 3 weeks
exceptions early enough during boot that the kernel will do ithe same.
Use lwsync only when compiling for LP64 and revert to the more proven isync
when compiling for ILP32. Note that in the end (i.e. between revision 222198
and this change) ILP32 changed from using sync to using isync. As per Nathan
the isync is needed to make sure I/O accesses are properly serialized with
locks and isync tends to be more effecient than sync.
While here, undefine __ATOMIC_ACQ and __ATOMIC_REL at the end of the file
so as not to leak their definitions.
Discussed with: nwhitehorn
- Use isync/lwsync unconditionally for acquire/release. Use of isync
guarantees a complete memory barrier, which is important for serialization
of bus space accesses with mutexes on multi-processor systems.
- Go back to using sync as the I/O memory barrier, which solves the same
problem as above with respect to mutex release using lwsync, while not
penalizing non-I/O operations like a return to sync on the atomic release
operations would.
- Place an acquisition barrier around thread lock acquisition in
cpu_switchin().
not provide general barriers, but only barriers in the context of the
atomic sequences here. As such, make them private and keep the global
*mb() routines using a variant of sync.
isync to implement read and write barriers, following Appendix B.2 of
Book II of the architecture manual. This provides a 25% speed increase
to fork() on the PowerPC G4.
of sync (lwsync is an alternate encoding of sync on systems that do not
support it, providing graceful fallback). This provides more than an order
of magnitude reduction in the time required to acquire or release a mutex.
MFC after: 2 months
Rewrite atomic operations for powerpc in order to achieve the following:
- Produce a type-clean implementation (in terms of functions arguments
and returned values) for the primitives.
- Fix errors with _long() atomics where they ended up with the wrong
arguments to be accepted.
- Follow the sys/type.h specifics that define the numbered types starting
from standard C types.
- Let _ptr() version to not auto-magically cast arguments, but leave
the burden on callers, as _ptr() atomic is intended to be used
relatively rarely.
Fix cfi in order to support the latest point.
In collabouration with: bde
Tested by: andreast, nwhitehorn, jceel
MFC after: 2 weeks
Kernel sources for 64-bit PowerPC, along with build-system changes to keep
32-bit kernels compiling (build system changes for 64-bit kernels are
coming later). Existing 32-bit PowerPC kernel configurations must be
updated after this change to specify their architecture.
and ifnet functions
- add memory barriers to <machine/atomic.h>
- update drivers to only conditionally define their own
- add lockless producer / consumer ring buffer
- remove ring buffer implementation from cxgb and update its callers
- add if_transmit(struct ifnet *ifp, struct mbuf *m) to ifnet to
allow drivers to efficiently manage multiple hardware queues
(i.e. not serialize all packets through one ifq)
- expose if_qflush to allow drivers to flush any driver managed queues
This work was supported by Bitgravity Inc. and Chelsio Inc.
so that all implemented variants have proper prototypes. The 8-bit,
16-bit and 64-bit variants are not implemented.
This really fixes the current build breakages caused by type casting
and struct aliasing rules.
cast as uint32_t which is defined as unsigned int. gcc doesn't want to
consider that there might not be much difference between an int and
a long on a 32 bit architecture.
variable and returns the previous value of the variable.
Tested on: i386, alpha, sparc64, arm (cognet)
Reviewed by: arch@
Submitted by: cognet (arm)
MFC after: 1 week
constraint is actually only allowed for register operands. Instead, use
separate input and output memory constraints.
Education from: alc
Reviewed by: alc
Tested on: i386, alpha
MFC after: 1 week
variables rather than void * variables. This makes it easier and simpler
to get asm constraints and volatile keywords correct.
MFC after: 3 days
Tested on: i386, alpha, sparc64
Compiled on: ia64, powerpc, amd64
Kernel toolchain busted on: arm
place.
This moves the dependency on GCC's and other compiler's features into
the central sys/cdefs.h file, while the individual source files can
then refer to #ifdef __COMPILER_FEATURE_FOO where they by now used to
refer to #if __GNUC__ > 3.1415 && __BARC__ <= 42.
By now, GCC and ICC (the Intel compiler) have been actively tested on
IA32 platforms by netchild. Extension to other compilers is supposed
to be possible, of course.
Submitted by: netchild
Reviewed by: various developers on arch@, some time ago
